Sleeping Threadpool: worker thread awakens without notify() from main thread - multithreading

I am implementing a thread pool where workers sleep when there is no work ready, and the main thread sleeps when workers are busy.
I noticed the worker threads are proceeding to work after calling wait() even though the main thread did not notify_all().
The output looks something like below:
WORKER WAIT
MAIN SETUP
MAIN POLLS READINESS
WORKER WAIT
WORKER WAIT
WORKER AWAKENS!
WORKER START
ALL WORKERS READY
WORKER AWAKENS!
......
Worker function:
void TaskSystemParallelThreadPoolSleeping::waitFunc() {
std::unique_lock<std::mutex> lock(*this->mutex_);
while(true) {
this->num_wait++;
std::cout << "WORKER WAIT" << std::endl;
this->cond_->wait(lock,
std::bind(&TaskSystemParallelThreadPoolSleeping::wakeWorker, this));
std::cout << "WORKER AWAKENS!" << std::endl;
if (this->done_flag == true) {
this->mutex_->unlock();
break;
}
this->mutex_->unlock();
std::cout << "WORKER START" << std::endl;
while (true) {
this->mutex_->lock();
if (this->not_done == 0) { // ALL work done
if (this->total_work != 0) { // 1st time seen by workers
this->total_work = 0;
this->num_wait = 0;
std::cout << "WORKER WAKE MAIN" << std::endl;
this->mutex_->unlock();
this->cond_->notify_all();
}
this->mutex_->unlock();
break;
}
int total = this->total_work;
int id = this->work_counter;
if (id == total) { // NO work initiated or NO work left
this->mutex_->unlock();
continue;
}
++(this->work_counter); // increment counter
this->mutex_->unlock(); // Let others access counters to work
this->runnable->runTask(id, total); // do work
this->mutex_->lock();
--(this->not_done); // decrement counter after work done
this->mutex_->unlock();
}
std::cout << "WORKER DONE" << std::endl;
}
std::cout << "WORKER TERMINATE" << std::endl;
}
Main thread:
void TaskSystemParallelThreadPoolSleeping::run(IRunnable* runnable, int num_total_tasks) {
//
// TODO: CS149 students will modify the implementation of this
// method in Part A. The implementation provided below runs all
// tasks sequentially on the calling thread.
// Set-up work
this->mutex_->lock();
std::cout << "MAIN SETUP" << std::endl;
this->runnable = runnable;
this->work_counter = 0;
this->not_done = num_total_tasks;
this->total_work = num_total_tasks;
// Tell workers there is work
std::cout << "MAIN POLLS READINESS" << std::endl;
while (this->num_wait < this->num_T) { // Check if all ready
this->mutex_->unlock();
this->mutex_->lock();
}
std::cout << "ALL WORKERS READY" << std::endl;
this->mutex_->unlock();
this->cond_->notify_all();
// Wait for workers to complete work
std::unique_lock<std::mutex> lock(*this->mutex_);
this->cond_->wait(lock,
std::bind(&TaskSystemParallelThreadPoolSleeping::wakeMain, this));
std::cout << "MAIN END" << std::endl;
}
Condition to wake worker:
bool TaskSystemParallelThreadPoolSleeping::wakeWorker() {
return (this->done_flag == true ||
(this->total_work != 0 && this->num_wait == this->num_T));
}
Condition to wake main thread:
bool TaskSystemParallelThreadPoolSleeping::wakeMain() {
return this->total_work == 0;
}
Thread Pool Constructor:
TaskSystemParallelThreadPoolSleeping::TaskSystemParallelThreadPoolSleeping(int num_threads): ITaskSystem(num_threads) {
//
// TODO: CS149 student implementations may decide to perform setup
// operations (such as thread pool construction) here.
// Implementations are free to add new class member variables
// (requiring changes to tasksys.h).
//
this->num_T = std::max(1, num_threads - 1);
this->threads = new std::thread[this->num_T];
this->mutex_ = new std::mutex();
this->cond_ = new std::condition_variable();
this->total_work = 0;
this->not_done = 0;
this->work_counter = 0;
this->num_wait = 0;
this->done_flag = {false};
for (int i = 0; i < this->num_T; i++) {
this->threads[i] = std::thread(&TaskSystemParallelThreadPoolSleeping::waitFunc, this);
}
}
Thread pool destructor:
TaskSystemParallelThreadPoolSleeping::~TaskSystemParallelThreadPoolSleeping() {
this->done_flag = true;
this->cond_->notify_all();
for (int i = 0; i < this->num_T; i++) {
this->threads[i].join();
}
delete this->mutex_;
delete[] this->threads;
delete this->cond_;
}
I think the beginning should be:
WORKER WAIT
MAIN SETUP
MAIN POLLS READINESS
WORKER WAIT
WORKER WAIT
ALL WORKERS READY
WORKER AWAKENS!
WORKER START
WORKER AWAKENS!
......
Ie. workers should only awaken after main's notify_all()
EDIT:
Here's the full log. It seems this self-awakening of workers causes a deadlock later, where one of the workers awakens by itself and does all the work, setting this->num_wait=0 and this->total_work=0. Therefore all the threads only see this->num_wait=1.
WORKER WAIT
MAIN SETUP
MAIN POLLS READINESS
WORKER WAIT
WORKER WAIT
WORKER AWAKENS!
WORKER START
ALL WORKERS READY
WORKER AWAKENS!
WORKER START
WORKER AWAKENS!
WORKER START
WORKER WAKE MAIN
WORKER DONE
WORKER WAIT
MAIN ENDWORKER DONE
WORKER WAIT
WORKER DONE
WORKER WAIT
MAIN SETUP
MAIN POLLS READINESS
ALL WORKERS READY
WORKER AWAKENS!
WORKER START
WORKER AWAKENS!
WORKER AWAKENS!
WORKER START
WORKER START
WORKER WAKE MAIN
WORKER DONE
WORKER WAIT
WORKER DONEMAIN END
MAIN SETUP
MAIN POLLS READINESS
WORKER DONE
WORKER WAIT
WORKER WAIT
WORKER AWAKENS!
WORKER START
ALL WORKERS READY
WORKER AWAKENS!
WORKER START
WORKER AWAKENS!
WORKER START
WORKER WAKE MAIN
WORKER DONEWORKER DONE
WORKER WAIT
WORKER WAIT
WORKER DONE
WORKER WAIT
MAIN END
MAIN SETUP
MAIN POLLS READINESS
ALL WORKERS READY
WORKER AWAKENS!
WORKER START
WORKER AWAKENS!
WORKER START
WORKER AWAKENS!
WORKER START
WORKER WAKE MAIN
WORKER DONE
WORKER WAIT
WORKER DONEWORKER DONE
WORKER WAIT
MAIN END
MAIN SETUP
MAIN POLLS READINESS
WORKER WAIT
WORKER AWAKENS!
WORKER START
WORKER WAKE MAIN
WORKER DONE
WORKER WAIT
ALL WORKERS READY
MAIN END
MAIN SETUP
MAIN POLLS READINESS

The reason for "worker thread awakens without notify from main thread" is pretty self-evident: the increment of this->num_wait and this->cond_->wait is within the same block without giving up the lock/notifying the main thread if condition variable wake up condition is true. The last thread in the poll passes the condition defined in wakeWorker() straightaway, hence your observation.
(I hope this code is just some toy code to play with -- it has too many issues ... it doesn't surprise me if it gets deadlocked given the manual mutex lock/unlock ...)
this->num_wait++;
std::cout << "WORKER WAIT" << std::endl;
this->cond_->wait(lock,
std::bind(&TaskSystemParallelThreadPoolSleeping::wakeWorker, this));
std::cout << "WORKER AWAKENS!" << std::endl;
According to https://en.cppreference.com/w/cpp/thread/condition_variable/wait, condition_variable::wait() is equivalent to
while (!pred()) {
wait(lock);
}
so if pred() return true, the execution doesn't give up lock

Related

Why is SIGUSR1 signal caught by epoll_wait instead of sigwait even though SIGUSR1 is blocked

I have a C++ program with a number of threads. At the start of the main thread (before any threads are spawned) I block SIGUSR1 using sigprocmask. I then create two threads, one waits for a SIGUSR1 using sigwait and the other monitors file descriptors using epoll_wait. For some reason when I send a SIGUSR1 to this program it is caught by the epoll_wait and not by the sigwait.
Linux Version 4.9.165
I have tried explicitly blocking the SIGUSR1 before calling epoll_wait but that does not solve the problem. I have also tried using epoll_pwait directly.
// main thread
// ...
sigset_t sigset;
sigemptyset(&sigset);
sigaddset(&sigset, SIGUSR1);
pthread_sigmask(SIG_SETMASK, &sigset, NULL);
// ...
//thread 1
// ...
sigset_t sigset;
int sig;
sigemptyset(&sigset);
sigaddset(&sigset, SIGUSR1);
auto ret = sigwait(&sigset, &sig);
if (ret != 0)
{
std::cout << "sigwait failed\n";
}
// ...
//thread 2
static constexpr int MAX_EVENTS = 10;
struct epoll_event events[MAX_EVENTS];
int timeout = -1;
// ...
sigset_t oldset;
sigprocmask(SIG_SETMASK, NULL, &oldset);
std::cout << (sigismember(&oldset, SIGUSR1) ? "SIGUSR1 blocked " : "SIGUSR1 not blocked ") << "in epoll thread\n";
epoll_wait(epollFd.Get(), static_cast<struct epoll_event*>(events), MAX_EVENTS, timeout);
if (nfds < 0)
{
std::cout << "Epoll wait failed: " << strerror(errno) << '\n';
}
// ...
This gives me the output "Epoll wait failed: Interrupted system call" an the sigwait thread never wakes up.
I also see the "SIGUSR1 blocked in epoll thread" indicating that before the epoll_wait the SIGUSR1 signal was blocked.
My understanding is that child threads should inherit the signal masks of their parents, that the epoll_wait should be treated as an epoll_pwait with a null sigmask (and thus not alter the signal mask), and that the signal should go to the only thread that has the signal unblocked (being the one with sigwait). Am I incorrect?

Deadlock after notify_one() (notifying thread locks mutex)

I have following problem. I have some multiple threads that do some work and one main thread that wakes them up when work is available. So far, I have managed to write some code using conditional variables and mutexes and most of the time this works okay, but from time to time, notifying thread will lock the mutex right after call notify_one(), thus blocking the notified thread and deadlocking.
I have written minimal code to illustrate this situation.
#include <iostream>
#include <thread>
#include <condition_variable>
std::mutex lock;
std::condition_variable cv;
void foo() {
std::cout << "Thread: Entering doWork()" << std::endl;
std::unique_lock<std::mutex> l(lock);
std::cout << "Thread: Acquired lock, going to wait." << std::endl;
cv.wait(l , []{return true;});
std::cout << "Thread: Done waiting, exit." << std::endl;
}
int main(void) {
std::unique_lock<std::mutex> l(lock);
std::cout << "MAIN: Creating thread." << std::endl;
std::thread t(foo);
std::cout << "MAIN: Unlocking mutex." << std::endl;
l.unlock();
std::cout << "MAIN: Notifying thread." << std::endl;
cv.notify_one();
//std::this_thread::sleep_for(std::chrono::seconds(1));
l.lock();
std::cout << "MAIN: Acquired lock." << std::endl;
std::cout << "MAIN: Joining thread." << std::endl;
t.join();
return 0;
}
In ideal situation, the output should be
MAIN: Creating thread.
MAIN: Unlocking mutex.
Thread: Entering doWork()
Thread: Acquired lock, going to wait.
MAIN: Notifying thread.
Thread: Done waiting, exit.
MAIN: Acquired lock.
MAIN: Joining thread.
but more often than not it is
MAIN: Creating thread.
MAIN: Unlocking mutex.
MAIN: Notifying thread.
MAIN: Acquired lock.
MAIN: Joining thread.
Thread: Entering doWork()
Is there any better way to eliminate chance of deadlock except of adding sleep into notifying thread (which I don't want to do)? Thank you in advance.
It's "condition variable" not "conditional variable", and the reason it's called that is that you use it to wait on some condition.
You aren't doing that, you just wait on a lambda that always returns true, and that's the cause of your problem. That and the fact your main thread is holding the lock all the time (why?!)
Sometimes the main thread runs the unlock, the notify_one, and the lock quickly, before the foo thread has even started. That means the foo thread misses the notification, then tries to acquire the lock, but can't because the main thread has it.
A condition variable is not like a semaphore, the notify_one` call does not set a state that can be detected later. If the condition variable isn't waiting when the notify_one call happens then it misses it, and it is gone forever. If you miss the notification and then sleep you will never wake up.
The solution is not to add arbitrary sleeps, that doesn't solve anything (ever!)
The correct solution is to have a condition that is tested, and to stop holding the lock when you're not updating any shared data. In the example below the condition being tested is "is the boolean ready true?" and the foo thread will wait until that condition is true. The main thread sets the variable, making the condition true, and then notifies the other thread that it should re-check the condition.
#include <iostream>
#include <thread>
#include <condition_variable>
std::mutex lock;
std::condition_variable cv;
bool ready = false;
void foo() {
std::cout << "Thread: Entering doWork()" << std::endl;
std::unique_lock<std::mutex> l(lock);
std::cout << "Thread: Acquired lock, going to wait." << std::endl;
cv.wait(l , []{return ready;});
std::cout << "Thread: Done waiting, exit." << std::endl;
}
int main(void) {
std::cout << "MAIN: Creating thread." << std::endl;
std::thread t(foo);
{
std::cout << "MAIN: Locking mutex." << std::endl;
std::unique_lock<std::mutex> l(lock);
ready = true;
}
std::cout << "MAIN: Notifying thread." << std::endl;
cv.notify_one();
std::cout << "MAIN: Joining thread." << std::endl;
t.join();
}

How to send signal/data from a worker thread to main thread?

I'll preface this by saying that I'm delving into multithreading for the first time. Despite a lot of reading on concurrency and synchronization, I'm not readily seeing a solution for the requirements I've been given.
Using C++11 and Boost, I'm trying to figure out how to send data from a worker thread to a main thread. The worker thread is spawned at the start of the application and continuously monitors a lock free queue. Objects populate this queue at various intervals. This part is working.
Once the data is available, it needs to be processed by the main thread since another signal will be sent to the rest of the application which cannot be on a worker thread. This is what I'm having trouble with.
If I have to block the main thread through a mutex or a condition variable until the worker thread is done, how will that improve responsiveness? I might as well just stay with a single thread so I have access to the data. I must be missing something here.
I have posted a couple questions, thinking that Boost::Asio was the way to go. There is an example of how signals and data can be sent between threads, but as the responses indicate, things get quickly overly-complicated and it's not working perfectly:
How to connect signal to boost::asio::io_service when posting work on different thread?
Boost::Asio with Main/Workers threads - Can I start event loop before posting work?
After speaking with some colleagues, it was suggested that two queues be used -- one input, one output. This would be in shared space and the output queue would be populated by the worker thread. The worker thread is always going but there would need to be a Timer, probably at the application level, that would force the main thread to examine the output queue to see if there were any pending tasks.
Any ideas on where I should direct my attention? Are there any techniques or strategies that might work for what I'm trying to do? I'll be looking at Timers next.
Thanks.
Edit: This is production code for a plugin system that post-processes simulation results. We are using C++11 first wherever possible, followed by Boost. We are using Boost's lockfree::queue. The application is doing what we want on a single thread but now we are trying to optimize where we see that there are performance issues (in this case, a calculation happening through another library). The main thread has a lot of responsibilities, including database access, which is why I want to limit what the worker thread actually does.
Update: I have already been successful in using std::thread to launch a worker thread that examines a Boost lock::free queue and processes tasks placed it in. It's step 5 in #Pressacco's response that I'm having trouble with. Any examples returning a value to the main thread when a worker thread is finished and informing the main thread, rather than simply waiting for the worker to finish?
If your objective is develop the solution from scratch (using native threads, queues, etc.):
create a thread save queue queue (Mutex/CriticalSection around add/remove)
create a counting semaphore that is associated with the queue
have one or more worker threads wait on the counting semaphore (i.e. the thread will block)
the semaphore is more efficient than having the thread constantly poll the queue
as messages/jobs are added to the queue, increment the semaphore
a thread will wake up
the thread should remove one message
if a result needs to be returned...
setup another: Queue+Semaphore+WorkerThreads
ADDITIONAL NOTES
If you decide to implement a thread safe queue from scratch, take a look at:
Synchronization between threads using Critical Section
With that said, I would take another look at BOOST. I haven't used the library, but from what I hear it will most likely contain some relevant data structures (e.g. a thread safe queue).
My favorite quote from the MSDN:
"When you use multithreading of any sort, you potentially expose
yourself to very serious and complex bugs"
SIDEBAR
Since you are looking at concurrent programming for the first time, you may wish to consider:
Is your objective to build production worthy code , or is this simply a learning exercise?
production? consider us existing proven libraries
learning? consider writing the code from scratch
Consider using a thread pool with an asynchronous callback instead of native threads.
more threads != better
Are threads really needed?
Follow the KISS principle.
The feedback above led me in the right direction for what I needed. The solution was definitely simpler than having to use signals/slots or Boost::Asio as I had previously attempted. I have two lock-free queues, one for input (on a worker thread) and one for output (on the main thread, populated by the worker thread). I use a timer to schedule when the output queue is processed. The code is below; perhaps it is of use to somebody:
//Task.h
#include <iostream>
#include <thread>
class Task
{
public:
Task(bool shutdown = false) : _shutdown(shutdown) {};
virtual ~Task() {};
bool IsShutdownRequest() { return _shutdown; }
virtual int Execute() = 0;
private:
bool _shutdown;
};
class ShutdownTask : public Task
{
public:
ShutdownTask() : Task(true) {}
virtual int Execute() { return -1; }
};
class TimeSeriesTask : public Task
{
public:
TimeSeriesTask(int value) : _value(value) {};
virtual int Execute()
{
std::cout << "Calculating on thread " << std::this_thread::get_id() << std::endl;
return _value * 2;
}
private:
int _value;
};
// Main.cpp : Defines the entry point for the console application.
#include "stdafx.h"
#include "afxwin.h"
#include <boost/lockfree/spsc_queue.hpp>
#include "Task.h"
static UINT_PTR ProcessDataCheckTimerID = 0;
static const int ProcessDataCheckPeriodInMilliseconds = 100;
class Manager
{
public:
Manager()
{
//Worker Thread with application lifetime that processes a lock free queue
_workerThread = std::thread(&Manager::ProcessInputData, this);
};
virtual ~Manager()
{
_workerThread.join();
};
void QueueData(int x)
{
if (x > 0)
{
_inputQueue.push(std::make_shared<TimeSeriesTask>(x));
}
else
{
_inputQueue.push(std::make_shared<ShutdownTask>());
}
}
void ProcessOutputData()
{
//process output data on the Main Thread
_outputQueue.consume_one([&](int value)
{
if (value < 0)
{
PostQuitMessage(WM_QUIT);
}
else
{
int result = value - 1;
std::cout << "Final result is " << result << " on thread " << std::this_thread::get_id() << std::endl;
}
});
}
private:
void ProcessInputData()
{
bool shutdown = false;
//Worker Thread processes input data indefinitely
do
{
_inputQueue.consume_one([&](std::shared_ptr<Task> task)
{
std::cout << "Getting element from input queue on thread " << std::this_thread::get_id() << std::endl;
if (task->IsShutdownRequest()) { shutdown = true; }
int result = task->Execute();
_outputQueue.push(result);
});
} while (shutdown == false);
}
std::thread _workerThread;
boost::lockfree::spsc_queue<std::shared_ptr<Task>, boost::lockfree::capacity<1024>> _inputQueue;
boost::lockfree::spsc_queue<int, boost::lockfree::capacity<1024>> _outputQueue;
};
std::shared_ptr<Manager> g_pMgr;
//timer to force Main Thread to process Manager's output queue
void CALLBACK TimerCallback(HWND hWnd, UINT nMsg, UINT nIDEvent, DWORD dwTime)
{
if (nIDEvent == ProcessDataCheckTimerID)
{
KillTimer(NULL, ProcessDataCheckPeriodInMilliseconds);
ProcessDataCheckTimerID = 0;
//call function to process data
g_pMgr->ProcessOutputData();
//reset timer
ProcessDataCheckTimerID = SetTimer(NULL, ProcessDataCheckTimerID, ProcessDataCheckPeriodInMilliseconds, (TIMERPROC)&TimerCallback);
}
}
int main()
{
std::cout << "Main thread is " << std::this_thread::get_id() << std::endl;
g_pMgr = std::make_shared<Manager>();
ProcessDataCheckTimerID = SetTimer(NULL, ProcessDataCheckTimerID, ProcessDataCheckPeriodInMilliseconds, (TIMERPROC)&TimerCallback);
//queue up some dummy data
for (int i = 1; i <= 10; i++)
{
g_pMgr->QueueData(i);
}
//queue a shutdown request
g_pMgr->QueueData(-1);
//fake the application's message loop
MSG msg;
bool shutdown = false;
while (shutdown == false)
{
if (GetMessage(&msg, NULL, 0, 0))
{
TranslateMessage(&msg);
DispatchMessage(&msg);
}
else
{
shutdown = true;
}
}
return 0;
}

Cancelling callbacks in Boost ASIO

I've been trying to switch my code from using one io_service per network connection to using shared ones, and I'm seeing some very odd behaviour on server sockets (client ones seem to work OK).
In order to try to work out what's going on I've re-started building up a simple example that will allow me to check my assumptions about everything that ought to happen. The first problem I've hit is that io_service::run doesn't exit when there are no handlers left, and as far as I can tell the handlers aren't removed from the work queue.
I have one thread that does an async_accept followed by an async_read. There is a separate client thread (which has its own io_service). The client thread's io_service is never run, and the server's one is run in yet another thread.
I'm using a condition variable to wait in the server thread for the read to complete (which will never happen as the client never writes). This times out just fine and then I call socket.cancel(). I would expect this to remove the read handler and run to exit as the work queue is now empty.
I do see the read handler get called (with a cancel error), but run never exits. When I tie the socket lifetime to the handler lifetime (by lambda capturing a shared_ptr to the socket) the memory isn't freed either.
The server is set up like this:
std::mutex mutex;
std::unique_lock<std::mutex> lock(mutex);
std::condition_variable signal;
boost::asio::io_service server_service;
boost::asio::ip::tcp::acceptor listener(server_service);
std::mutex read_mutex;
std::unique_lock<std::mutex> read_lock(read_mutex);
std::condition_variable read_done;
std::thread server([&]() {
std::unique_lock<std::mutex> lock(mutex);
listener.open(boost::asio::ip::tcp::v4());
listener.set_option(boost::asio::socket_base::enable_connection_aborted(true));
listener.bind(boost::asio::ip::tcp::endpoint(boost::asio::ip::tcp::v4(), 4567));
listener.listen();
std::shared_ptr<connection> server_cnx(new connection(server_service));
listener.async_accept(server_cnx->socket,
[&, server_cnx](const boost::system::error_code& error) {
log_thread() << "Server got a connection " << error << std::endl;
boost::asio::async_read_until(server_cnx->socket, server_cnx->buffer, '\n',
[&, server_cnx](const boost::system::error_code& error, std::size_t bytes) {
log_thread() << "Got " << bytes << ", " << error << std::endl;
std::unique_lock<std::mutex> lock(read_mutex);
lock.unlock();
read_done.notify_one();
});
});
lock.unlock();
signal.notify_one();
if ( read_done.wait_for(read_lock, std::chrono::seconds(1)) == std::cv_status::timeout ) {
log_thread() << "Server read timed out -- cancelling socket jobs" << std::endl;
server_cnx->socket.cancel();
server_cnx->socket.close();
} else {
log_thread() << "Server data read" << std::endl;
}
log_thread() << "Exiting server thread" << std::endl;
});
signal.wait(lock);
log_thread() << "Server set up" << std::endl;
The io_service thread is set up like this:
std::thread server_io([&]() {
log_thread() << "About to service server IO requests" << std::endl;
try {
server_service.run();
} catch ( ... ) {
log_thread() << "Exception caught" << std::endl;
}
log_thread() << "**** Service jobs all run" << std::endl;
signal.notify_one();
});
The output is as follows:
10.0002 139992957945728 Server set up
10.0005 139992957945728 Client set up
10.0006 139992848398080 About to service server IO requests
10.0006 139992848398080 Server got a connection system:0
11.0003 139992934819584 Server read timed out -- cancelling socket jobs
11.0004 139992934819584 Exiting server thread
11.0004 139992848398080 Got 0, system:125
20.0006 139992957945728 IO thread timed out servicing requests -- stopping it
^^^ This should not happen because the server service should have run out of work
20.0006 139992957945728 Waiting for things to close....
22.0008 139992957945728 Wait over, exiting
(Columns are time + 10s, thread ID, log message)
At the 11 second mark you can see that the async_read_until is called. This is the last handler in the server's io_service and yet run doesn't exit.
Even after the time out waiting for run to exit fires and the waiting thread does io_service::stop(), still run doesn't exit (there's another 2 second wait there).
The full code is on github
The program is invoking undefined behavior when the server thread attempts to unlock the read_lock that it does not own.
int main()
{
...
std::mutex read_mutex;
std::unique_lock<std::mutex> read_lock(read_mutex); // Acquired by main.
std::condition_variable read_done;
std::thread server([&]() { // Capture lock reference.
std::unique_lock<std::mutex> lock(mutex);
...
// The next line invokes undefined behavior as this thread does did
// not acquire read_lock.mutex().
if (read_done.wait_for(read_lock, ...)
// ^^^^^^^^^ caller does not own.
{
...
}
});
signal.wait(lock);
...
}
In particular, when invoking condition_variable::wait_for(lock), the standard requires that lock.owns_lock() is true and lock.mutex() is locked by the calling thread.
Mixing synchronous and asynchronous flows often add complexity. In this particular case, where the synchronous calls are intertwined throughout each layer using lower-level constructs for event/signal notification without a persisted state, I think that it adds unnecessary complexity and overcomplicates the flow. Furthermore, the broad scope of variables can add complexity. If read_lock had never been captured by the lambdas, then a compiler error would have occurred.
Consider the separation in space when trying to observe two events:
// I will eventually be interested when the server starts
// accepting connections, so start setting up now.
std::mutex server_mutex;
std::unique_lock<std::mutex> server_lock(server_mutex);
std::condition_variable server_started;
std::thread server([&]()
{
// I will eventually be interested when the server reads
// data, so start setting up now.
std::mutex read_mutex;
std::unique_lock<std::mutex> read_lock(read_mutex);
std::condition_variable read_done;
listener.async_accept(...,
[&](...)
{
// Got connection.
async_read_until(...,
[&](...)
{
// Someone may be interested that data has been read,
// so use the correct mutex and condition_variable
// pair.
std::unique_lock<std::mutex> read_lock(read_mutex);
read_lock.unlock();
read_done.notify_one();
});
}); // async_accept
// Someone may be interested that I am accepting connections,
// so use the correct mutex and condition_variable pair.
std::unique_lock<std::mutex> server_lock(server_mutex);
server_lock.unlock();
server_done.notify_one();
// I am now interested in if data has been read.
read_done.wait_for(read_lock);
}); // server thread
// I am now interested in if the server has started.
server_started.wait(server_lock);
The caller has to prepare to handle an event, start an operation, then wait for the event, and the operation must know the event the caller is interested in. To worsen the situation, one must now consider lock ordering to prevent deadlocks. Note how in the above example, the server thread acquires the read_mutex and then the server_mutex. Another thread cannot acquire the mutexes in a difference order without introducing the chance of a deadlock. In terms of complexity, this approach scales poorly with the number of events.
It may be worth considering re-examining the program's flow and control structure. If it can be written to be primarily asynchronous, then callback chains, continuations, or a signal-and-slot system (Boost.Signals) may uncomplicate the solution. If one prefers to have asynchronous code read as if it was synchronous, then Boost.Asio's support for coroutines can provide a clean solution. Finally, if one needs to synchronously wait on an asynchronous operation's result or timeout, then consider using Boost.Asio's support for std::future or using them directly.
// Use an asynchronous operation so that it can be cancelled on timeout.
std::future<std::size_t> on_read = boost::asio::async_read_until(
socket, buffer, '\n',boost::asio::use_future);
// If timeout occurs, then cancel the operation.
if (on_read.wait_for(std::chrono::seconds(1)) == std::future_status::timeout)
{
socket.cancel();
}
// Otherwise, the operation completed (with success or error).
else
{
// If the operation failed, then on_read.get() will throw a
// boost::system::system_error.
auto bytes_transferred = on_read.get();
}
While I would strongly advocate re-examining the overall control structure and reducing variable scope, the following sample is roughly equivalent to the above example, but may be slightly easier to maintain with its use of std::future:
// I will eventually be interested when the server starts
// accepting connections, so start setting up now.
std::promise<void> server_started_promise;
auto server_started = server_started_promise.get_future();
std::thread server([&]()
{
// I will eventually be interested when the server reads
// data, so start setting up now.
std::promise<void> read_done_promise;
auto read_done = read_done_promise.get_future();
listener.async_accept(...,
[&](...)
{
// Got connection.
async_read_until(...,
[&](...)
{
// Someone may be interested that data has been read.
read_done_promise.set_value();
});
}); // async_accept
// Someone may be interested that I am accepting connections.
server_started_promise.set_value();
// I am now interested in if data has been read.
read_done.wait_for(...);
}); // server thread
// I am now interested in if the server has started.
server_started.wait();
Here is a complete example based on the original code that demonstrates using std::future to control flow and timeout asynchronous operations in a synchronous manner:
#include <future>
#include <iostream>
#include <thread>
#include <boost/asio.hpp>
#include <boost/asio/use_future.hpp>
#include <boost/optional.hpp>
#include <boost/utility/in_place_factory.hpp>
int main()
{
using boost::asio::ip::tcp;
// Setup server thread.
boost::asio::io_service server_io_service;
std::promise<tcp::endpoint> server_promise;
auto server_future = server_promise.get_future();
// Start server thread.
std::thread server_thread(
[&server_io_service, &server_promise]
{
tcp::acceptor acceptor(server_io_service);
acceptor.open(tcp::v4());
acceptor.set_option(
boost::asio::socket_base::enable_connection_aborted(true));
acceptor.bind(tcp::endpoint(tcp::v4(), 0));
acceptor.listen();
// Handlers will not chain work, so control the io_service with a work
// object.
boost::optional<boost::asio::io_service::work> work(
boost::in_place(std::ref(server_io_service)));
// Accept a connection.
tcp::socket server_socket(server_io_service);
auto on_accept = acceptor.async_accept(server_socket,
boost::asio::use_future);
// Server has started, so notify caller.
server_promise.set_value(acceptor.local_endpoint());
// Wait for connection or error.
boost::system::system_error error =
make_error_code(boost::system::errc::success);
try
{
on_accept.get();
}
catch (const boost::system::system_error& e)
{
error = e;
}
std::cout << "Server got a connection " << error.code() << std::endl;
// Read from connection.
boost::asio::streambuf buffer;
auto on_read = boost::asio::async_read_until(
server_socket, buffer, '\n', boost::asio::use_future);
// The async_read operation is work, so destroy the work object allowing
// run() to exit.
work = boost::none;
// Timeout the async read operation.
if (on_read.wait_for(std::chrono::seconds(1)) ==
std::future_status::timeout)
{
std::cout << "Server read timed out -- cancelling socket jobs"
<< std::endl;
server_socket.close();
}
else
{
error = make_error_code(boost::system::errc::success);
std::size_t bytes_transferred = 0;
try
{
bytes_transferred = on_read.get();
}
catch (const boost::system::system_error& e)
{
error = e;
}
std::cout << "Got " << bytes_transferred << ", "
<< error.code() << std::endl;
}
std::cout << "Exiting server thread" << std::endl;
});
// Wait for server to start accepting connections.
auto server_endpoint = server_future.get();
std::cout << "Server set up" << std::endl;
// Client thread.
std::promise<void> promise;
auto future = promise.get_future();
std::thread client_thread(
[&server_endpoint, &promise]
{
boost::asio::io_service io_service;
tcp::socket client_socket(io_service);
boost::system::error_code error;
client_socket.connect(server_endpoint, error);
std::cout << "Connected " << error << std::endl;
promise.set_value();
// Keep client socket alive, allowing server to timeout.
std::this_thread::sleep_for(std::chrono::seconds(2));
std::cout << "Exiting client thread" << std::endl;
});
// Wait for client to connect.
future.get();
std::cout << "Client set up" << std::endl;
// Reset generic promise and future.
promise = std::promise<void>();
future = promise.get_future();
// Run server's io_service.
std::thread server_io_thread(
[&server_io_service, &promise]
{
std::cout << "About to service server IO requests" << std::endl;
try
{
server_io_service.run();
}
catch (const std::exception& e)
{
std::cout << "Exception caught: " << e.what() << std::endl;
}
std::cout << "Service jobs all run" << std::endl;
promise.set_value();
});
if (future.wait_for(std::chrono::seconds(3)) ==
std::future_status::timeout)
{
std::cout << "IO thread timed out servicing requests -- stopping it"
<< std::endl;
server_io_service.stop();
}
// Join all threads.
server_io_thread.join();
server_thread.join();
client_thread.join();
}
So of course this multi-threading is a tricky business. Turns out that in this case the read lock is acquired in the wrong place so the handler is being blocked by the thread waiting for it to complete.
I guess the lesson here is to never handle thread locks without some sort of time out.

Thread management in a game loop?

I am in the middle of developing a game and came across the problem of multithreading. I already used multithreading successfully when loading resources. I did that by creating some threads at some point, assigned them functions, and waited for them to finish, while drawing a loading screen, pretty straightforward.
Now I want to create some threads, that can wait idle till they receive a function, when they do, solve that, then stay idle again. They must operate in a game loop, which is roughly like this (I came up with these function names just for easy visualization):
std::thread t0,t1;
while(gamerunning)
{
UpdateGame();
t0.receiveFunc( RenderShadow );
t1.receiveFunc( RenderScene );
WaitForThreadstoFinishWork();
RenderEverything(); //Only draw everything if the threads finished (D3D11's Deferred Context rendering)
}
t0.Destroy();
t1.Destroy();
My rendering engine is working, and for the time being (for testing), I created threads in my game loop, which is a terrible way of even a quick test, because my rendering speed even slowed down. By the way, I am using C++11's library.
Long story short, I want to create threads before my game loop takes place, and use those in the game loop afterwards, hope someone can help me out. If it is an option, I would really want to stay away from the lower levels of threading, I just need the most straightforward way of doing this.
Following your most recent comments, here is an example implementation of a thread that wakes up on demand, runs its corresponding task and then goes back to sleep, along with the necessary functions to manage it (wait for task completion, ask for shutdown, wait for shutdown).
Since your set of functions is fixed, all you'll have left to do is to create as much threads as you need (ie. 7, probably in a vector), each with its own corresponding task.
Note that once you remove the debugging couts there's little code left, so I don't think there is a need to explain the code (it's pretty self-explanatory IMHO). However don't hesitate to ask if you need explanations on some details.
class TaskThread {
public:
TaskThread(std::function<void ()> task)
: m_task(std::move(task)),
m_wakeup(false),
m_stop(false),
m_thread(&TaskThread::taskFunc, this)
{}
~TaskThread() { stop(); join(); }
// wake up the thread and execute the task
void wakeup() {
auto lock = std::unique_lock<std::mutex>(m_wakemutex);
std::cout << "main: sending wakeup signal..." << std::endl;
m_wakeup = true;
m_wakecond.notify_one();
}
// wait for the task to complete
void wait() {
auto lock = std::unique_lock<std::mutex>(m_waitmutex);
std::cout << "main: waiting for task completion..." << std::endl;
while (m_wakeup)
m_waitcond.wait(lock);
std::cout << "main: task completed!" << std::endl;
}
// ask the thread to stop
void stop() {
auto lock = std::unique_lock<std::mutex>(m_wakemutex);
std::cout << "main: sending stop signal..." << std::endl;
m_stop = true;
m_wakecond.notify_one();
}
// wait for the thread to actually be stopped
void join() {
std::cout << "main: waiting for join..." << std::endl;
m_thread.join();
std::cout << "main: joined!" << std::endl;
}
private:
std::function<void ()> m_task;
// wake up the thread
std::atomic<bool> m_wakeup;
bool m_stop;
std::mutex m_wakemutex;
std::condition_variable m_wakecond;
// wait for the thread to finish its task
std::mutex m_waitmutex;
std::condition_variable m_waitcond;
std::thread m_thread;
void taskFunc() {
while (true) {
{
auto lock = std::unique_lock<std::mutex>(m_wakemutex);
std::cout << "thread: waiting for wakeup or stop signal..." << std::endl;
while (!m_wakeup && !m_stop)
m_wakecond.wait(lock);
if (m_stop) {
std::cout << "thread: got stop signal!" << std::endl;
return;
}
std::cout << "thread: got wakeup signal!" << std::endl;
}
std::cout << "thread: running the task..." << std::endl;
// you should probably do something cleaner than catch (...)
// just ensure that no exception propagates from m_task() to taskFunc()
try { m_task(); } catch (...) {}
std::cout << "thread: task completed!" << std::endl;
std::cout << "thread: sending task completed signal..." << std::endl;
// m_wakeup is atomic so there is no concurrency issue with wait()
m_wakeup = false;
m_waitcond.notify_all();
}
}
};
int main()
{
// example thread, you should really make a pool (eg. vector<TaskThread>)
TaskThread thread([]() { std::cout << "task: running!" << std::endl; });
for (int i = 0; i < 2; ++i) { // dummy example loop
thread.wakeup();
// wake up other threads in your thread pool
thread.wait();
// wait for other threads in your thread pool
}
}
Here's what I get (actual order varies from run to run depending on thread scheduling):
main: sending wakeup signal...
main: waiting for task completion...
thread: waiting for wakeup or stop signal...
thread: got wakeup signal!
thread: running the task...
task: running!
thread: task completed!
thread: sending task completed signal...
thread: waiting for wakeup or stop signal...
main: task completed!
main: sending wakeup signal...
main: waiting for task completion...
thread: got wakeup signal!
thread: running the task...
task: running!
thread: task completed!
thread: sending task completed signal...
thread: waiting for wakeup or stop signal...
main: task completed!
main: sending stop signal...
main: waiting for join...
thread: got stop signal!
main: joined!

Resources